PaperlayerView exact revision on arXiv

Source-linked AI summary

Characterizing LLM-Based Family Education through the Lens of Activity Theory: A Scoping Review of the HCI Literature

Lan Luo, Yuqi Liang, Jie Cai, Anqi Wang, Dongyijie Pan, Muzhi Zhou, Chun Yu, Pan Hui

arXiv:2609.28886v1cs.HC

TL;DR

HCI lacks a systematic account of the educational interactions emerging around LLMs in family education. This review applies an activity-theoretical framework to 53 studies, finding a focus on child–parent interaction and a small set of educational areas, while LLM capabilities take meaning through situated use and redistribute educational labour.

  • Problem

    HCI research has not systematically explained LLM-based family education or how its distinct interactions relate across systems and educational contexts.

    Source: §1 · p. 2

  • Method

    The review examines 53 HCI studies using an activity-theoretical framework to relate educational objects and participation to mediation, labour, and rules.

    Source: §1 · p. 2, §6 · p. 27

  • Results

    Research centers mainly on child–parent interaction and language, narrative, AI literacy, and relational development, while LLM capabilities acquire educational meaning through situated use.

    Source: §6 · p. 27

  • Takeaways & Limitations

    The framework explains how LLM capabilities become organized through family participation and how their use redistributes educational labour while leaving parents responsible for much supervision.

    Source: §6 · p. 27

    Reported distributions characterize the retrieved corpus rather than the prevalence of family education.

    Source: §5.5 · p. 26

Abstract

from arXiv · show

Large language models (LLMs) are increasingly involved in family education, yet HCI has not systematically explained the educational interactions that emerge around them. This scoping review analyzes 53 HCI studies from 6,540 records across 19 venues. Using activity theory and AODM, it relates participants and educational objects to mediation, labour, and rules. We find that the literature centers on child--parent interaction and on language, AI literacy, and relational learning. The introduction of LLMs enabled conversational, embodied, and spatial systems to generate support from the context of an unfolding interaction. LLMs redistributed educational labour, while family and institutional rules left parents and professionals responsible for interpreting outputs and deciding how they entered practice. Evidence across families and educational purposes remains limited, especially on sustained personalization, repair labour, and how families negotiate authority and rules. The review offers a framework explaining how LLM capabilities become organized through family participation.

1 Introduction

HCI research has described distinct ways LLMs enter family learning, but lacks a shared account of how these interactions relate. This review applies activity theory through AODM to synthesize 53 studies and identify research gaps and HCI directions.

1 Introduction

  • Family education includes children’s learning with family members and parental learning within family life.The framing includes learning across generations as well as parents developing and revising their understanding of parenting.Source: §1 · pp. 1–2
  • Prior work spans shared stories and questions [15], systems turning child-activity interpretations into parent guidance [17], and homeschooling decisions shaped by curricular expectations and family values [81].These examples range from immediate learning tasks to how families interpret and organize education together.Source: §1 · p. 2
  • Existing HCI research lacks a systematic account of LLM-based family education and a shared framework for relating the interactions studied across systems and contexts.The review identifies this gap as the motivation for its synthesis.Source: §1 · p. 2
  • The review uses AODM to identify each activity’s educational object, examine LLM mediation, and relate system roles to participation and responsibility.Its classification also considers who participates, system capabilities and implementation, divisions of labour, and rules.Source: §1 · p. 2
  • The 53-study synthesis examines educational objects and outcomes, participants and labour, mediation of relationships, and household, institutional, platform, and cultural rules.These are the review’s four areas of analysis.Source: §1 · p. 2
  • The review contributes a systematic synthesis, research gaps derived from evidence limitations and tensions, and an HCI agenda on family representation, capability alignment, labour, and negotiable rules.The agenda addresses negotiation across household and external contexts.Source: §1 · pp. 2–3

2 Related Work

This section situates family education as cooperative participation, reviews earlier technology strands and the gap in cross-study synthesis of educational LLM interactions, and presents activity theory and AODM as lenses for examining how systems mediate educational objects and organize participation.

2 Related Work

  • Related work frames the need to synthesize how family members use LLM-based systems for educational goals and introduces activity theory operationalized through AODM as the analytical lens.Source: §2 · p. 3

2.1.1 Family Education Through Cooperative Interaction.

  • Family education is cooperative and interdependent, encompassing shared learning, participation in another member’s learning, and relationships involving children or adult family members.Source: §2.1.1 · p. 3

2.1 Family Education Through Cooperative Interaction Using LLMs

  • LLMs can generate educational support as family interaction unfolds, and this review examines family members’ engagement with LLM-based systems in educational interaction.Source: §2.1.1 · p. 3, §2.1.2 · pp. 3–4

2.1.2 Earlier Technological Support before LLMs in Family Education.

  • Earlier HCI support spanned conversational, physically embodied, and spatial or immersive systems; conversational systems commonly used designer-specified dialogue spaces, mapping speech to predefined intents and returning prewritten responses.Source: §2.1.2 · p. 4

2.2 Prior Reviews of Technology in Family and Educational Contexts

  • Prior reviews cover family media use, child-centered AI, and educational generative AI, but family participation remains peripheral and HCI lacks a cross-study account of family interactions with educational LLMs.Source: §2.2 · pp. 4–5

2.3 Activity Theory as a Lens for LLM-based Family Education

  • Activity theory treats technology as a mediating tool directed toward an educational object and organized through collective participation, situating boundary coordination within the activity.Source: §2.3 · p. 5
  • Rules structure how participants may act, divisions of labor distribute educational responsibility, and community includes groups beyond the immediate interaction who share the educational object.Figure 1 presents these activity-system components and their relationships in LLM-based family education.Source: §2.3 · p. 5
  • AODM translates activity-system components and relations into research questions and representations for empirical analysis.The review uses it to analyze how LLM systems mediate educational objects and how rules, communities, and divisions of labor organize participation.Source: §2.3 · p. 5
  • Prior AODM applications analyzed coordinated healthcare practices and technology-mediated learning activities pursuing shared educational objects.In this review, AODM structures analysis across the corpus.Source: §2.3 · p. 5

3 Scoping Review Method

The review maps HCI research on LLM-based family education through a scoping process, using activity theory and AODM to guide analysis of a screened 53-study corpus. Its synthesis compares family educational activities and how LLM use is organized, distinguishing family use from other forms of LLM involvement.

3 Scoping Review Method

  • The review used Activity Theory and AODM to chart how LLM systems mediate family educational activities and organize participation around educational goals.It mapped patterns by research question and identified gaps arising from limitations or fragmentation in the evidence.Source: §3 · p. 6

3.1 Research Questions

  • The review adapted AODM’s activity-system components into questions about participants, educational objects and outcomes, technical mediation, division of labor, and rules.The questions covered family members and external communities, system capabilities and implementations, and how participation and LLM use are organized.Source: §3.1 · p. 6

3.2 Search Strategy and Record Identification

  • Searches across selected HCI venues identified 6,540 records, and deduplication left 6,455 unique records for title-and-abstract screening.The search used metadata and a Boolean query spanning LLM technologies, education, and family contexts; no lower publication-date bound was imposed.Source: §3.2 · pp. 7–8

3.3 Study Selection and Eligibility Assessment

  • Two authors developed preliminary eligibility criteria from the review scope and research questions, then independently screened a shared pilot set before formal screening.The supplied passage introduces title-and-abstract screening of the 6,455 unique records.Source: §3.3.1 · p. 8, §3.3.2 · p. 9

3.3.1 Criteria Development and Calibration.

  • In the 50-record calibration exercise, the authors agreed on 49 records (98% raw agreement; Cohen’s κ = 0.960) before discussing the disagreement and refining criteria.These statistics describe calibration, not the subsequent formal screening.Source: §3.3.1 · p. 8
  • Eligibility required primary empirical evidence, family-organized participation, an explicit educational objective, and substantive LLM involvement.Exclusions were applied in order, from secondary research or absent empirical evidence to absent educational aims or substantive LLM use, then incidental family context.Source: §3.3.1 · pp. 8–9

3.3.2 Screening Process and Results.

  • Title-and-abstract screening produced 213 agreed inclusions (99.77% raw agreement; Cohen’s κ = 0.965) and excluded 6,242 records.Exclusions were 258 under EC1, 188 under EC2, 3,763 under EC3, 16 under EC4, and 2,017 under EC5.Source: §3.3.2 · p. 9
  • Eligible studies demonstrated substantive LLM involvement in educational interactions organized through family relationships; completed empirical evidence qualified regardless of publication length or contribution type, and researcher use to analyze family educational interactions also qualified [34].Source: §3.3.2 · p. 9
  • Full-text screening retained 53 studies: researchers agreed on 53 inclusions (99.53% raw agreement; Cohen’s κ = 0.988) after excluding 160 reports.The 160 exclusions were 47 under EC1, 23 under EC2, 67 under EC3, 3 under EC4, and 20 under EC5; the remaining studies satisfied IC1–IC4.Source: §3.3.2 · p. 9

3.4 Data Analysis

  • AODM defined the activity components, and each study’s focal configuration was identified from its central question, design goal, or empirical objective and checked against its description, design, findings, and discussion.When a study described multiple configurations, the one most aligned with its central objective was focal; other configurations remained contextual information.Source: §3.4 · pp. 9–10
  • Study-level records captured activity components, and researchers calibrated the codebook on 10 studies before dividing the remaining 43 studies among three coders.Rules required extra discussion because studies did not always distinguish system controls from family or institutional expectations.Source: §3.4 · p. 10
  • The synthesis separates systems used in family activity, support mediated by family members or professionals, and researcher tools; dataset representation alone does not show family LLM use or educational change.Reported evidence included learning or interaction findings, perceptions, design requirements, and technical results; requirements and computational accuracy alone do not demonstrate educational change.Source: §3.4 · p. 10

4 Findings

The findings describe a 53-study corpus and organize analysis around the activity-system relations represented in the research questions. The corpus spans multiple publication years, venues, and study forms.

4 Findings

  • The findings first describe the corpus, then examine activity-system relations among subjects, communities, educational objects, mediation, labour, and rules.Source: §4 · p. 10

4.1 Corpus Overview

  • The corpus contains 53 studies selected from 6,540 records retrieved across 19 HCI venues.Source: §4.1 · p. 10
  • Publication counts rose sharply after 2023, from one study each in 2022 and 2023 to 10 in 2024, 23 in 2025, and 18 in 2026.Source: §4.1 · p. 10
  • The studies appeared in 10 venues, led by CHI with 30 and IDC with nine, and included system evaluations, qualitative investigations, design explorations, and computational work.Source: §4.1 · pp. 10–11

4.2 Findings 1: Participants, Educational Objects, and Outcomes in LLM-Based Family Education

This section examines the participant configurations, educational objects, and outcomes represented in the reviewed studies of LLM-based family education. It describes the range of learning aims and evaluated outcomes while noting limited coverage across family circumstances and learning domains, and the rare connection of individual learning, family interaction, and technical performance.

4.2.1 Child–Parent Relationships and Selective External Expertise (RQ1.1).

  • Child–parent configurations accounted for 44 of 53 studies (83%); the most-addressed age bands were 5–8 (29 studies) and 7–11 (25).Age bands overlap, and the counts include intended users in parent-focused research.Source: §4.2.1 · p. 11, Table 1 · p. 11, Table 2 · p. 12
  • External expertise entered when family activity required knowledge or authority beyond the household, linking learning with educational goals, clinical and care support, and research or parenting guidance.Examples included teachers and homeschooling networks, speech-language pathologists and clinicians, and research or parenting communities.Source: §4.2.1 · p. 12
  • Family configurations beyond child–parent activity and diverse family circumstances remain underrepresented, limiting evidence about who guides LLM use and whose experiences findings represent.The corpus included one grandparent–grandchild study and none centered siblings; demographic characteristics and family resources were reported inconsistently.Source: §4.2.1 · p. 13

4.2.2 Educational Objects Concentrated on Language, Interpretation, and Family Support (RQ1.2).

  • The six educational-object categories describe intended learning purposes or study foci, not demonstrated gains; language and narrative work included vocabulary, personalized stories, co-reading, and expressive communication.Studies also addressed storytelling across generations and distance, spatial language during block play, and family interpretation of stories or artwork.Source: §4.2.2 · p. 13
  • Other educational objects included spatial problem-solving and perspective interpretation, emotional and relational development, and communication across generations, distance, and care contexts.Activities included block play, comparing perspectives and critiquing stories, emotion discussion, and communication in healthcare or developmental support.Source: §4.2.2 · p. 13
  • AI and digital literacy made the technology itself a learning object, while self-directed learning concerned children's growing independence and parents' mediation of AI use.Families considered responsible use, chatbot and video appropriateness, trust, educational expectations, and family values.Source: §4.2.2 · pp. 13–14
  • Mathematical, scientific, and embodied learning were weakly represented, and longitudinal evidence on how educational objects develop through family routines remained limited.Several studies examined bounded activities, while Culture Link studied family storytelling in a ten-day field trial.Source: §4.2.2 · pp. 14–15
  • Educational-object studies examined collaborative content evaluation, video appropriateness, self-directed learning and parental involvement, and homeschooling curriculum development in Homeroom.These studies begin to connect LLM use with family organization; longitudinal research is needed to trace how educational objects, child development, and family support shape one another across recurring routines.Source: §4.2.2 · p. 15

4.2.3 Reported Outcomes and Their Evidence (RQ1.3).

  • Children’s evaluations covered spatial language, vocabulary, story interpretation, emotional identification, narrative expression, and dialogic-reading engagement; self-directed learning elicited parental perspectives, not demonstrated readiness gains.These measures concern different aspects of participation and development.Source: §4.2.3 · p. 15
  • Studies examined parent–child communication strategies, home speech-practice needs, special-education navigation, and intergenerational learning about responsible GenAI use.Desired or co-designed support establishes what parents need, whereas claims of improved parental knowledge require evidence of learning.Source: §4.2.3 · p. 15
  • Research examined family dialogue and turn-taking, shared media reflection, intergenerational storytelling, and design or appraisal of systems and materials, without establishing lasting educational effects.These findings span different stages of design and use.Source: §4.2.3 · p. 15
  • Studies elicited media-appropriateness judgments, moderation preferences, autonomy views, household-agent expectations, values, and trust or acceptance perceptions, not evidence of more effective LLM use.These findings inform design and adoption rather than demonstrating that families learned to use LLMs more effectively.Source: §4.2.3 · p. 15
  • AACessTalk documented parent–child turn-taking [17], eaSEL shared media reflection, and co-creation grandparent–grandchild connection [54], but studies rarely linked these findings with technical performance and learning.The review calls for evaluating system performance alongside family interaction and educational outcomes within the same study.Source: §4.2.3 · p. 16

4.3 Findings 2: How LLM-Based Systems Mediate Family Education

The review finds that family-facing LLM systems mediate educational activity through interfaces and participant arrangements, while retrospective studies use LLMs to analyze interaction records. Across systems, generated support takes different roles according to context and family participation.

4.3 Findings 2: How LLM-Based Systems Mediate Family Education

  • Family-facing systems mediate educational objects through interfaces and participant arrangements, while retrospective studies use LLMs to analyze interaction records; four recurring forms were materials, dialogue, context, and external knowledge or values.Source: §4.3 · p. 16

4.3.1 LLM Contributions Across Interface Forms.

  • Conversational systems use preceding turns and participant input to generate material within unfolding exchanges, whose educational role depends on how people interpret and continue them.Source: §4.3.1 · p. 17
  • Embodied interaction arranges generation alongside adult participation, while material and spatial interfaces use physical artifacts as context for generated contributions.Source: §4.3.1 · p. 17

4.3.2 Mediation Functions and Educational Objects.

  • Generated materials draw on family context for stories, co-creation, and curricula, becoming provisional resources whose relevance is completed through family participation.Source: §4.3.2 · pp. 18–19
  • Systems structure dialogue through questions, contrasting perspectives, prompts, and responses, with children contributing interpretations and adults extending exchanges.Source: §4.3.2 · pp. 18–19
  • Systems interpret family activity for guidance or decision-making and connect external resources such as IEP knowledge, pediatric communication, curriculum standards, and family-planning knowledge.Source: §4.3.2 · p. 19
  • Personalization most often adapts support to information available at a particular interaction, with less evidence on changing learner representations, repairing mistaken profiles, or negotiating their use.Source: §4.3.2 · p. 19
  • Evaluations of LLM-generated analyses established analytic or benchmark performance, but evidence remains needed on whether families or professionals understand outputs and use them in decisions and practice.Source: §4.3.2 · p. 19

4.4 Findings 3: How Participation and LLM Use Are Organized

These findings examine how LLM roles and family, platform, and institutional rules organize participation in family education, including confidentiality, responsibility, and accountability obligations. They also note limited evidence about sustained use, labor, negotiation, and conflicts where these rules meet.

4.4.1 LLM Roles and Human Responsibilities (RQ3.1).

  • Task allocations identify responsibilities, not net workload change; studies often expected parents to review or adapt outputs but documented detection, correction, and failure-explanation labor less consistently [44].Evidence about who checks outputs, corrects errors, coordinates participation, and manages emotional consequences is needed to document redistribution.Source: §4.4.1 · pp. 20–21
  • Systems interacted directly with learners through reading material [15], dialogic prompts, and emotional-reflection support, or supplied resources for parents to interpret and adapt [17] [63].Home speech-practice research elicited desired AI support through parent interviews and expert review of concepts.Source: §4.4.1 · p. 20
  • In shared family activity, members connected generated material with their experiences and negotiated its meaning [54], while other systems linked family concerns with professional expertise [70].The child-welfare analysis concerned institutional information work but did not establish how families used its outputs.Source: §4.4.1 · pp. 20–21
  • Labor redistribution has mainly been examined in parent–child configurations, with much thinner evidence for siblings, multiple caregivers, extended families, and changing participation across households.Comparative research could examine whether LLM use concentrates work in one caregiver or shifts responsibility among family members and households.Source: §4.4.1 · p. 21
  • Responsibility across family and professional boundaries remains unresolved: research should distinguish professional review from family contextualization and trace who interprets, approves, contests, and remains accountable for guidance.Community-connected systems brought clinical and educational knowledge into family activity, while child-welfare analysis raised institutional accountability questions [70].Source: §4.4.1 · p. 21

4.4.2 Reported Rules for Authority, Acceptability, and Information Boundaries (RQ3.2).

  • Participation rules coordinated child autonomy and adult authority, as parents could adjust involvement in children’s educational-robot learning [44]; acceptability linked system participation to situated judgment, with reliability, emotional safety, and value alignment recurring concerns.Source: §4.4.2 · p. 23
  • Information boundaries joined household privacy with external accountability: disclosure and parental access intersected with documentation, confidentiality, and institutional obligations [23] [70].Household preferences, platform practices, and external obligations jointly shaped those boundaries.Source: §4.4.2 · p. 23
  • Anticipated rules are documented more often than rules enacted or revised during sustained use, and evidence is limited on how families resolve tensions between child privacy and agency, parental oversight, and external obligations.Repeated-use studies could examine how relationships, educational purposes, and children’s growing experience change initially acceptable rules.Source: §4.4.2 · p. 23
  • Clinical systems brought confidentiality and responsibility into exchanges involving children and parents, while public-service applications placed family information within institutional documentation and accountability practices [70].Source: §4.4.2 · p. 24
  • The corpus describes these obligations separately and provides little evidence about conflicts with household values or platform rules.Source: §4.4.2 · p. 24
  • Studying LLM use where household and institutional rules meet could reveal whose interpretation governs disputed actions, how families and professionals contest decisions, and who remains accountable for consequences.Source: §4.4.2 · p. 24

5 Discussion

The discussion uses activity theory to examine how participation, mediation, labor, rules, and decision authority shape generation’s educational role, and frames related design questions. It also notes that the review’s conclusions are bounded by primary-study reporting and analytical choices, so its design directions are interpretations to investigate.

5.1 Rethinking the Boundaries of LLM-Based Family Education

  • 44 of 53 studies centered child–parent relationships; one centered grandparents and grandchildren, none siblings, and objects clustered around language, narrative, AI literacy, and relational development.Logical reasoning [63] and self-directed learning received less attention; future systems could preserve whose contributions shape activities and connect generated dialogue with physical exploration.Source: §5.1 · p. 24

5.2 Evaluating Long-Term Personalization Across Changing Family Educational Needs

  • The reviewed personalization offers limited evidence that adaptation stays appropriate as learners develop and family purposes change.Longitudinal evaluation should examine outdated profiles, differing child and caregiver interpretations, how disagreements are resolved, and whether revisions align support with changing goals.Source: §5.2 · p. 25

5.3 Rethinking Assistance as a Redistribution of Family Labor

  • Systems take on generation and preliminary analysis, but parents still interpret BrickSmart guidance [63], fit speech-practice support to routines, and review Homeroom materials [81], so net workload reduction is unproven.The key question is what guidance parents contribute and when they intervene, not conversational volume alone.Source: §5.3 · p. 25
  • Evaluations should measure review time, corrections, handoffs, and acceptance or rejection of guidance alongside usability and learning outcomes, since the balance of effort saved and introduced is unknown.Generation may save preparation while requiring additional checking, correction, and coordination.Source: §5.3 · p. 25
  • When guidance crosses household and professional boundaries, systems should preserve sources and distinguish generated interpretations from authoritative decisions.Explicit review states could show who considered a recommendation, what remains disputed, and who can authorize action.Source: §5.3 · pp. 25–26

5.4 From Configurable Controls to Negotiated Family Governance

  • The corpus describes expectations about access, supervision, content, authority, and sharing, but these are anticipated rules more than rules enacted and revised during sustained use.Systems could represent rules as revisable agreements, tracking who proposed them, their scope, and when to reconsider them.Source: §5.4 · p. 26
  • Selective disclosure should let children preview sharing, caregivers explain access requests, and families contest inferences or correct records without assuming comprehensive parental visibility suits every activity.Designs should distinguish routine monitoring from safety escalation and document unresolved disagreement.Source: §5.4 · p. 26

5.5 Limitations of the Review

  • The venue-based search and metadata terms may omit relevant work, so corpus distributions describe retrieved studies rather than LLM use in families or the full family-education literature.The corpus combines evaluated systems, needs and expectations investigations, and retrospective analyses; desired features and model performance do not establish observed benefits or changes in family learning.Source: §5.5 · p. 26
  • The review’s conclusions depend on what primary studies reported about participants, practices, and evaluation.Source: §5.5 · p. 26
  • Representing each study through a focal activity configuration aids comparison but can underrepresent secondary relationships and changes within an activity.Source: §5.5 · p. 26
  • AODM foregrounds relations among participants, tools, purposes, and responsibilities, while other analytical frameworks could expose different patterns.Source: §5.5 · p. 26
  • The proposed design directions should be treated as interpretations to investigate.Source: §5.5 · p. 27

6 Conclusion

The review uses activity theory and AODM to connect family participants and educational objects with the technologies, labour, and rules organizing activity. It finds that current studies cluster around child–parent interaction and a limited set of learning areas, while LLM use redistributes educational labour and leaves important evidence gaps.

6 Conclusion

  • The review examines 53 HCI studies using an activity-theoretical framework operationalized through AODM to connect participants and objects with tools, labour, and rules.The analysis relates family participants and educational objects to the technologies and social organization of each activity.Source: §6 · p. 27
  • Studies center mainly on child–parent interaction and language, narrative, AI literacy, and relational development, with other relationships, domains, and sustained routines less represented.The available evidence is also limited by short deployments.Source: §6 · p. 27
  • LLM capabilities acquire educational meaning through implementation and situated use, as systems contribute support that family members judge for fit with the learner and situation.Systems supply provisional materials, structure interaction, interpret activity, and connect households with external knowledge.Source: §6 · p. 27
  • LLM use redistributes educational labour while leaving parents responsible for much of the supervision and correction.The review also notes preservation of professional authority over clinical or institutional knowledge.Source: §6 · p. 27
  • Family and information rules determine what systems can contribute and who can act on their outputs.The framework points toward explaining how families negotiate participation, responsibility, and authority when LLMs enter educational activity.Source: §6 · p. 27
  • The evidence remains concentrated on child–parent dyads and short deployments; future work should study diverse family configurations over time and follow analyses into the decisions they inform.This can move HCI beyond evaluating individual systems toward explaining how families negotiate participation, responsibility, and authority.Source: §6 · p. 27

AI Use Disclosure

The authors disclose using OpenAI tools for limited editing and visualization assistance while retaining responsibility for the research and manuscript.

AI Use Disclosure

  • OpenAI’s GPT supported language editing, consistency checks, and drafting revisions to evidence descriptions and review limitations.Source: AI Use Disclosure · p. 27
  • The authors retained responsibility for study selection, coding, interpretation, and the final manuscript.Source: AI Use Disclosure · p. 27
  • OpenAI Codex assisted with Figure 2 based on concepts and intended meaning specified by the authors, who reviewed and approved the visualization.Source: AI Use Disclosure · p. 27

Appendix 1: Search Information and Coding Framework

The appendix documents the search platforms and query adaptations and introduces the AODM-based coding framework and evidence distinctions.

Appendix 1: Search Information and Coding Framework

Appendix 2: Paper-Level Coding

The paper-level coding appendix explains how findings and activity components are recorded across the 53-study corpus and clarifies how to interpret those records.

Appendix 2: Paper-Level Coding

Continued on next page

This section maps family-education and family-facing GenAI studies, summarizing their reported aims, findings, applications, and how tools, rules, educational labour, and evaluative responsibilities are organized across learning, communication, design, parental support, and health-information contexts.

Continued on next page

  • Technical and assessment-oriented entries describe diarization, transcription, conversational alignment, vocabulary profiling, assessment, and personalized intervention.Some entries report only proposed support, expert feedback, or technical evaluation rather than deployed-system learning outcomes.Source: Continued on next page · p. 36
  • Design and co-design entries identify needs around appropriate content, parental involvement, safety, creativity, accessible engagement, and parent participation in IEP processes.Other reported requirements include adaptable materials, feedback, agency, responsibility, and parent advocacy support.Source: Continued on next page · pp. 36–37
  • Reported outcomes span engagement, recall, participation, vocabulary learning, AI-literacy dialogue, and family understanding across educational activities.Other entries describe anti-bullying reflection and family participation, as well as question quality, verbal engagement, and reading participation.Source: Continued on next page · pp. 36–37
  • Several entries report family communication and relational outcomes, including emotional expression, shared storytelling, social bonding, and intergenerational cultural transmission.The table also notes agency tensions and parent-informed design requirements.Source: Continued on next page · pp. 36, 38, 39
  • Parental support findings include agency, aligned materials, routine support, trusted assistance, communication guidance, and improved caregiver-professional interaction.Some entries report parent understanding and communication support following psychological-state assessment.Source: Continued on next page · pp. 37–38
  • Entries on AI literacy and family AI use report critical awareness, responsible engagement, parental mediation, and communication about AI across generations.The table also includes design-fiction findings about imagined benefits, frictions, and support needs.Source: Continued on next page · pp. 38–39
  • In other cases, reports mention detection of progress, deviations, and substantive concerns, while explicitly noting that family learning was not demonstrated in the reported cases.Source: Continued on next page · p. 39
  • The listed cases connect household safety and reproductive-health understanding with design requirements for threat responses, boundaries, privacy, resilience assessment, and sustained support.The table also names information access, agency, reflective dialogue, and support needs.Source: Continued on next page · p. 39
  • P49 and P50 foreground reciprocal intergenerational learning and critical reflection, including digital stewardship, responsible GenAI use, stereotype awareness, respect, and pluralism.Their listed design concerns include sustained support and parent-informed reflection.Source: Continued on next page · p. 39
  • P13–P18 describe shared AI-literacy inquiry, parent- and professional-supported practice, family interpretation of generated material, and therapeutic facilitation.The entries include professional boundaries, accessibility, adult oversight, authorship, bias, misinformation, and emotionally appropriate use.Source: Continued on next page · p. 41
  • The passages mention IEP procedures and documentation, parents’ advocacy, and schools’ and professionals’ retained institutional authority, as well as age-appropriate use, safety, privacy, adolescent agency, and appropriate caregiver access.Source: Continued on next page · pp. 41–42
  • Dialogic systems and augmented toys structure children’s reading and anti-bullying activities, with parents supporting evaluation and participating in enactment.Adults also identify requirements and mediate prospective child use; developmental fit is a stated concern.Source: Continued on next page · p. 42
  • Family-facing systems redistribute support work: parents provide experiences or context, while caregivers record and interpret recommendations and coordinate care.Remote parents contribute experiences for generated bedtime stories; caregivers also receive and interpret health information, with disclosure boundaries shaping participation.Source: Continued on next page · pp. 42–43
  • Systems assess or generate support, but parents and professionals retain judgment over developmental fit, transparency, clinical safety, and appropriate representation.Examples include parents comparing system reasoning before final decisions, and care experts appraising a pediatric chatbot’s proposed guidance.Source: Continued on next page · p. 42
  • Parental mediation spans children’s readiness and video use, adolescents’ AI literacy, and families’ consideration of AI support for autistic children.Parents’ reported acceptance and intentions do not establish adoption; privacy, accuracy, reminders, and caregiver coordination are concerns.Source: Continued on next page · pp. 38, 42, 43
  • Child-facing and creative systems invite emotional expression, reflection, and co-creation, while positioning parents to evaluate risks and teach socially appropriate participation.Children identify and express emotions, create with peers and AI, or construct sand scenes; parents interpret generated content and experts may validate reports or consult.Source: Continued on next page · p. 43
  • Generative artifacts can prompt shared family storytelling, while adolescents interpret and circulate AI-generated social-media content with parents and teachers participating.The family-story system composes shared artifacts for joint interpretation and retelling; the adolescent study emphasizes youth agency, critical awareness, and responsible communication.Source: Continued on next page · pp. 43–44
  • Practitioner assessments inform case-record evaluation, while discretion, confidentiality, public accountability, and avoiding automation remain salient.The case-analysis approach involved researchers analyzing records and practitioner assessments.Source: Continued on next page · p. 44
  • Families considering GenAI interventions articulate expectations around household privacy, consent, authority, transparency, and differentiated responses; postpartum women assess ChatGPT information as family and health actors shape access.The postpartum context also raises autonomy and appropriate professional-boundary considerations.Source: Continued on next page · p. 44
  • A conversational agent supports reflective dialogue and developmentally appropriate assessment, with privacy and clinical-support boundaries shaping children’s responses.Adults provide the surrounding support context, with reciprocal respect and verification also identified.Source: Continued on next page · p. 44
  • In everyday consumer GenAI use, younger family members introduce and troubleshoot tools while older members contribute judgment and context, enabling reciprocal learning.Cultural fit, fairness, responsibility, and age-appropriate stewardship also figure in family learning about these tools.Source: Continued on next page · p. 44
  • In LLM-supported role-play, parents inform scenarios and the system poses reflective questions that help children examine and challenge stereotypes.Family values, pluralism, and child-related considerations shape this activity.Source: Continued on next page · p. 44
  • GenAI cultural probes prompt children and parents to reflect together, articulate family values, and set design boundaries while avoiding prescriptive beliefs.Access, perceived risk, and educational value also inform these reflections.Source: Continued on next page · p. 44
  • Parents evaluate whether generative AI systems should be used by children and shape decisions about their educational value and parental responsibility.The passage frames these systems as being considered for children’s use.Source: Continued on next page · p. 44
  • With ELLA, parents select vocabulary and themes while the robot generates stories and scaffolds developmentally appropriate dialogue for children’s learning.The design also emphasizes parent-selected targets and bounded daily use.Source: Continued on next page · p. 44
Loading 2609.28886v1…